CN104010029B - DCE performance prediction method based on laterally longitudinal information integration - Google Patents
DCE performance prediction method based on laterally longitudinal information integration Download PDFInfo
- Publication number
- CN104010029B CN104010029B CN201410198278.4A CN201410198278A CN104010029B CN 104010029 B CN104010029 B CN 104010029B CN 201410198278 A CN201410198278 A CN 201410198278A CN 104010029 B CN104010029 B CN 104010029B
- Authority
- CN
- China
- Prior art keywords
- prediction
- mrow
- data
- time
- longitudinal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of DCE performance prediction method based on laterally longitudinal information integration, first, is cut into a long time series data Time Sub-series of multiple equal lengths, the length of each Time Sub-series is T;Afterwards, it is predicted using longitudinal prediction algorithm using Exponential Smoothing curved line arithmetic and the Time Sub-series of all generations before is predicted, the relevance between cycle and cycle is calculated, longitudinal prediction data is obtained and obtained longitudinal prediction data is integrated;Finally, it is predicted using lateral prediction algorithm by relation pair longitudinal prediction data obtained in the previous step between two time points within the time cycle, find the contact of the data in the cycle, build relational model, the horizontal relationship existed in all time cycles of searching is stored, enters line period interior prediction, weighted superposition predicts the outcome, and horizontal relationship is adjusted, finally predicted the outcome.The present invention improves accuracy and reliability of the server to scheduling of resource.
Description
Technical field
It is more particularly to a kind of to be based on horizontal longitudinal algorithm to dividing the present invention relates to information integration, time series forecasting field
The Forecasting Methodology of server performance under cloth environment.
Background technology
, it is necessary to which the relevant information of various resources is to determine that resource whether may be used in acquisition system in the load balancing of server
With then dispatching algorithm determines the priority of task according to availability, run time of task of resource etc. and distributes to it
Available resource.However as the operation of task, the state of various resources, such as cpu load, free memory, hard disk are remaining empty
Between etc. can change at any time, it is therefore desirable to server performance prediction come guide service device load balancing with scheduling calculate
Method.
Time series algorithm can be taken to server performance prediction, that is, made prediction based on historical data.The party
The easy steps of method:1) periodically acquisition server performance data in temporal sequence;2) these historical datas are based on, one is set up
The individual relational model between server performance and time variable;3) calculated using this model corresponding to the specified time
Server performance value, and using the value as server performance predicted value., can be to server using such model
It can be predicted, so as to help scheduler program preferably to distribute resource, management role, improve the work effect of whole distributed system
Rate.
Server performance is predicted using the method for time series, can be with passage time sequence autoregression model AR
Model, moving average model MA models etc. are predicted, and these models mainly can be carried out accurately to the data of some stable states
Prediction, but if data are not in stable situation, predict the outcome just not ideal enough, and these models can not be good
Carry out long-term forecast.
Conventional research shows that server performance situation is existed periodically.One server performance can be counted as
How multiple small, with different cycles periodic synergistic effects, prediction is instructed using the periodicity of data, is one
The problem of letter is to be solved.
The content of the invention
There is provided the side that long-term forecast is carried out to server performance for deficiencies of the prior art by the present invention
Method.The present invention is achieved through the following technical solutions:
A kind of DCE performance prediction method based on laterally longitudinal information integration, comprises the following steps:
Step 1:One long time series data is cut into the Time Sub-series of multiple equal lengths, each Time Sub-series
Length be T;
Step 2:It is predicted using longitudinal prediction algorithm using Exponential Smoothing curved line arithmetic to all generations of step 1
Time Sub-series are predicted, and calculate the relevance between cycle and cycle, obtain longitudinal prediction data and vertical to what is obtained
Integrated to prediction data;
Step 3:Obtained using lateral prediction algorithm by the relation pair step 2 between two time points within the time cycle
Longitudinal prediction data be predicted, specifically find the contact of the data in the cycle, build relational model, to the institute of searching sometimes
Between cycle memory horizontal relationship store, then enter line period interior prediction, weighted superposition predicts the outcome, and pass through feedback
Mode adjusts horizontal relationship, is finally predicted the outcome.
It is pre- according to the DCE performance based on laterally longitudinal information integration described in present pre-ferred embodiments
Historical data is specially converted into multigroup Time Sub-series by survey method, step 1 according to period of time T, wherein:
Historical data is:{x1,x2,x3,...,xn, length is n;
The length of each Time Sub-series is T, is cut into n/T Time Sub-series, i-th of subsequence SiIt is expressed as { xi1,
xi2,xi3,…,xiT, then historical data is cut into<S1,S2,…,Si,…,SL>, wherein L=n/T:
Wherein, x1,x2,x3,...,xnIt is historical data, SiIt is Time Sub-series,
xi1,xi2,xi3,…,xiTThe data of Time Sub-series after cutting.
It is pre- according to the DCE performance based on laterally longitudinal information integration described in present pre-ferred embodiments
Survey method, step 2 comprises the following steps:
Step 21:Long-term forecast is carried out to the time series of all generations of the first step using Exponential Smoothing curved line arithmetic,
The length of prediction walks for T, predicts the outcome as follows:
Wherein,It is longitudinal prediction data;
Step 22:Consolidated forecast data, prediction number is formed by step 21 prediction data point according to the sequence combination of original sampled point
Be according to the direction of, each fallout predictor prediction it is longitudinal, data finally splicing be it is horizontal, such as
For horizontal prediction data, length is T.
It is pre- according to the DCE performance based on laterally longitudinal information integration described in present pre-ferred embodiments
Survey method, step 3 comprises the following steps:
Step 31:Calculate the horizontal relationship between same all midcycle datas
Horizontal relationship refers to the relation between two time points within the time cycle, specific as follows:
Assuming that two data points are x, y, definition y=bx+a is the horizontal association between x, y, wherein, a, b is parameter;
Assuming that two data points are x, y, definition y=bx+a is the horizontal association between x, y, wherein, a, b is parameter;
A, b parameter are directly obtained at linear regression, i.e., linear regression method least square method formula is as follows:
Q=(y1-bx1-a)2+(y2-bx2-a)2+…+(yn-bxn-a)2
Minimize Q, first derivation show that design factor formula is as follows:
Wherein, Q is predicted value and the error of calculated value;
It whether there is horizontal relationship for two data points, pass through the linear regression come out to cycle training at the same time
Parameter and checking, if its error is less than certain threshold values, then it is assumed that there is horizontal relation;Otherwise, if error exceedes threshold values
Then think horizontal association is not present before the two time points;
Step 32:The horizontal relationship existed in all time cycles is found, and is stored it;
Step 33:Predict the outcome weighted superposition
After the horizontal relationship of all presence has been searched for by the predicted value of the predicted value of longitudinal algorithm and horizontal algorithm with
Weighting scheme is superimposed, and obtains prediction weighted results as follows:
Wherein:
It is the prediction weighted results of the i-th row jth row;
w1With w2It is the weight of laterally and longitudinally algorithm respectively;
It is the result of horizontal algorithm prediction;
It is that longitudinal algorithm predicts the outcome;
Step 34:Data are adjusted by horizontal relationship
Horizontal relationship is applied not only to prediction, is also used for adjustment finally to data, it is assumed that two time points of A, B exist horizontal
It is y=ax+b to relation, wherein, A corresponds to x, and B corresponds to y, and a, b is known parameters, also, is obtained by prediction above
A, B prediction data are xp, yp, then,
Set up cost function:
Q=(y-yp)2+(x-xp)2=(ax+b-yp)2+(x-xp)2
It is exactly that two data points meet horizontal relationship that Cost function, which are minimized, carries out first derivation and is set to 0:
2a(aX+b-yp)+2(X-xp)=0
It is X, aX+b respectively to obtain new prediction data point A, B.
It is pre- according to the DCE performance based on laterally longitudinal information integration described in present pre-ferred embodiments
Survey method, in addition to:
Step 4:The horizontal longitudinal prediction algorithm of dynamic integrity
Correspondence different machines different time sections, cycle time relevance is not that fixed situation is laterally longitudinal using dynamic
Prediction algorithm, be specially:
Assuming that prediction length in future is relatively, suffered time cycle influence is the same, in each prediction, will be gone through
History data remove last identical prediction length, then call the laterally pre- measuring and calculating in longitudinal direction respectively with a series of candidate time cycle
Method is predicted and calculates its error, that minimum time cycle of error is used for as the parameter of laterally longitudinal prediction algorithm pre-
Survey, i.e., according to the different time cycles, laterally longitudinal prediction algorithm is become into multiple fallout predictors, according to last prediction knot
Really, current optimal time cycle corresponding fallout predictor is selected.
It is pre- according to the DCE performance based on laterally longitudinal information integration described in present pre-ferred embodiments
The candidate time cycle in survey method, step 4 is respectively 6 hours, 12 hours, 1 day and 1 week.
The characteristics of present invention is directed to server performance, is carried out long-term pre- by laterally longitudinal prediction algorithm to behavior pattern
Survey.By the load data of acquisition server, the fine-grained analysis of a performance data of server, multiple time sequences are cut into
There is a data cycle in column data, each time series data, carrying out data with reference to laterally and longitudinally algorithm carries out long-term forecast,
Last integrated one collect predict the outcome, improve accuracy and reliability of the server to scheduling of resource.
Brief description of the drawings
Fig. 1 is horizontal vertical structure principle schematic of the invention;
Fig. 2 is laterally longitudinal algorithm flow schematic diagram of the invention;
Fig. 3 is weighting algorithm schematic diagram of the invention.
Embodiment
Below with reference to the accompanying drawing of the present invention, clear, complete description is carried out to the technical scheme in the embodiment of the present invention
And discussion, it is clear that as described herein is only a part of example of the present invention, is not whole examples, based on the present invention
In embodiment, the every other implementation that those of ordinary skill in the art are obtained on the premise of creative work is not made
Example, belongs to protection scope of the present invention.
For the ease of the understanding to the embodiment of the present invention, make further by taking specific embodiment as an example below in conjunction with accompanying drawing
Illustrate, and each embodiment does not constitute the restriction to the embodiment of the present invention.
A kind of DCE performance prediction method based on laterally longitudinal information integration, a server performance
The fine-grained analysis of data, is cut into multiple time series datas, and each time series data has a data cycle, with reference to
Laterally and longitudinally algorithm to carry out data long-term forecast, last integrated one collect predict the outcome.See Fig. 1,2 schematic diagrames,
Wherein:
Lateral prediction algorithm:Carry out Analysis server performance data from horizontal angle, between the data for calculating a cycle
Relation, whole data belong to periodic, and adjacent data the former can have an impact to the latter, it is known that point is with future position same
Time cycle, can make prediction value.
Longitudinal prediction algorithm:The contact in multiple cycles is analyzed from regulation of longitudinal angle, associating between cycle and cycle is calculated
Data interact between property, cycle, so as to predict next cycle data according to cycle stage.
Concretely comprise the following steps:
S1:One long time series data is cut into the Time Sub-series of multiple equal lengths, each Time Sub-series
Length is T.
Historical data is specially converted into multigroup time series according to the time cycle (T).
Historical data is:{x1,x2,x3,...,xn, length is n;
The length of each Time Sub-series is T, then can be cut into n/T Time Sub-series, i-th of subsequence SiCan be with
It is expressed as { xi1,xi2,xi3,…,xiT, then historical data can be cut into<S1,S2,…,Si,…,SL>, wherein L=n/T:
Wherein x1,x2,x3,...,xnIt is historical data, SiIt is subsequence data,
xi1,xi2,xi3,…,xiTData after cutting.
For example:If the former loaded sampling period is that 1h is t=1h, and T=24h then generates 24 after the step
Time series cycle, the sampling period is 24h, i.e., daily 1 point, daily 2 points...Daily 24 points totally 24 time serieses.
Function Mapping is:
Input:{x1,x2,x3,..,xn}
Output:{{xi,xi+1,xi+2..., xi+T| i=1,2 ... T }
More direct form (1) carrys out tracking data.
Table 1 builds multigroup time series schematic diagram
On the selection of time cycle, be typically chosen 1 day or 1 week it is more meaningful because no matter machine or the mankind
Activity 1 day or the periodic associated property of 1 time-of-week it is stronger, loadtype is a specific type related to network, is also entered
One step confirm the time cycle select 1 day and 1 week be rational.The selection other times cycle is perhaps feasible, but simply and special
Determine machine relation, applicability is little.
S2, the son to all generations of step 1 is predicted using Exponential Smoothing curved line arithmetic using longitudinal prediction algorithm
Time series is predicted, and calculates the relevance between cycle and cycle, obtains longitudinal prediction data and the longitudinal direction to obtaining
Prediction data is integrated.
The prediction algorithm of longitudinal direction can select any one existing prediction algorithm.Because Scheme result in longitudinal direction
During prediction, the time series models each decomposed need to only predict seldom point, such as the legacy data sampling period is 1h (t=
1h), cycle time is set to 1 week, if each time series predicts a value, prediction length can reach 168 steps, thus
The algorithm for considering that the short-term forecast degree of accuracy the is higher it can be seen that algorithm that longitudinal direction prediction is selected should try one's best, so can be only achieved relatively good
Effect.This algorithm has selected the generally acknowledged accurate algorithm of short-term forecast --- Exponential Smoothing curved line arithmetic
(exponential smoothing).The algorithm has relatively good effect for the short-term forecast of any irregular data.
Specifically, step S2 comprises the following steps:
S21:Long-term forecast, prediction are carried out to the time series of all generations of the first step using Exponential Smoothing curved line arithmetic
Length be T step, predict the outcome as follows:
Wherein,It is longitudinal prediction data;
Because data are to exist periodically, there is correlation between the data of same position, such as the first of each cycle
Individual data are related, therefore carry out long-term forecast using longitudinal prediction algorithm, and the length of prediction can be T steps.
S22:Consolidated forecast data, prediction data is formed by step S21 prediction data point according to the sequence combination of original sampled point,
Each fallout predictor prediction direction be it is longitudinal, data finally splicing be it is horizontal, such as
For horizontal prediction data, length is T.
S3:Obtained using lateral prediction algorithm by the relation pair step 2 between two time points within the time cycle
Longitudinal prediction data is predicted, and specifically finds the contact of the data in the cycle, relational model is built, to all times of searching
Cycle memory horizontal relationship store, then enter line period interior prediction, weighted superposition predicts the outcome, and passes through feedback side
Formula adjusts horizontal relationship, is finally predicted the outcome.
S31:Calculate the horizontal relationship between same all midcycle datas
Horizontal relationship refers to the relation (pass at generally referred to as neighbouring time point within the time cycle between two time points
Connection).Association such as two time points is linear relationship, because data are the line of two discrete, adjacent data in itself
It is exactly a broken line, and linear model is more simpler than other models.So being defined as one with regard to horizontal association in this model
Linear relation is planted, it is specific as follows:
Assuming that two data points are x, y, definition y=bx+a is the horizontal association between x, y, wherein, a, b is parameter;
A, b parameter are directly obtained at linear regression, i.e., linear regression method least square method formula is as follows:
Q=(y1-bx1-a)2+(y2-bx2-a)2+…+(yn-bxn-a)2
Minimize Q, first derivation show that design factor formula is as follows:
Wherein, Q is predicted value and the error of calculated value;
It whether there is horizontal relationship for two data points, pass through the linear regression come out to cycle training at the same time
Parameter and checking, if its error is less than certain threshold values, then it is assumed that there is horizontal relation;Otherwise, if error exceedes threshold values
Then think horizontal association is not present before the two time points.
S32:The horizontal relationship existed in all time cycles is found, and is stored it.
S33:Predict the outcome weighted superposition
When having searched for the horizontal relationship of all presence, if length of history data is not the integer of just time cycle
Times, unnecessary time point can be for prediction., just can basis when there is horizontal relationship between over head time point and non-future position
Horizontal line association is predicted.Because the algorithm of longitudinal direction can predict a value, so the value of horizontal algorithm prediction can be with weighting side
Formula is superimposed, and specific algorithm is shown in Fig. 3.
WhereinIt is the prediction weighted results of the i-th row jth row;
w1With w2It is the weight of laterally and longitudinally algorithm respectively;
It is the result of horizontal algorithm prediction;
It is that longitudinal algorithm predicts the outcome.
It should be noted that length of history data is exactly the time cycle at that time or unnecessary time point is with being future position
It is inscrutable in the absence of horizontal relationship this method.This method can only make prediction to fractional prediction point.For example, add
From 1 day 0 July during historical data:On 00 to July 3 12:00, the setting time cycle is July at finally have more one section if 1 day
The 0 of 3 days:00 to 12:If 00 finds by historical data, 12:00 and 13:00 has certain association, then can pass through 12:00
Load estimation 13:00 load.If but search horizontal relationship in previous step and do not find 13:00 and before 0:00 to 12:00
In the presence of horizontal association, then 13:00 is to make prediction in this step, and specific algorithm is shown in Fig. 3.
S34:Data are adjusted by horizontal relationship
Horizontal relationship is applied not only to prediction, can also such as know historical record further to adjustment finally to data
Show daily 8:00 (x) and 8:There is y=x+1 relation in 01 (y), although data point x has been got well in prediction abovep,yp, still
Can adjust so that it meets this relation.The specific derivation of equation is as follows:
Assuming that two time points of A B (A corresponds to x, and B corresponds to that y) to there is horizontal relationship be y=ax+b (known to a, b), and
And obtain xp yp during A B prediction data by prediction above.
Set up cost function:
Q=(y-yp)2+(x-xp)2=(ax+b-yp)2+(x-xp)2
It is exactly that two data points meet this relation that Cost function, which are minimized, carries out first derivation and is set to 0:
2a(aX+b-yp)+2(X-xP)=0
So new prediction data point AB is X, aX+b respectively
It is this prediction can be not only used for two need predict data point it, known data point and unknown number can also be used
Between strong point, because all horizontal relationships found in the first step of lateral prediction can be used.
Furthermore, it is contemplated that different machines different time sections, cycle time relevance be not it is fixed, many times in other words
It is to be influenceed by multiple time cycles, so it is not a selection well to fix a time cycle, therefore, the present invention is also
, can be with the relatively good this point that avoids including the horizontal longitudinal prediction algorithm of dynamic, laterally longitudinal prediction algorithm hypothesis is adjacent for dynamic
The influence of suffered time cycle of prediction length is the same.
Specifically, present invention additionally comprises following steps:
S4:The horizontal longitudinal prediction algorithm of dynamic integrity
Correspondence different machines different time sections, cycle time relevance is not that fixed situation is laterally longitudinal using dynamic
Prediction algorithm, be specially:
Assuming that prediction length in future is relatively, suffered time cycle influence is the same, in each prediction, will be gone through
History data remove last identical prediction length, then call the laterally pre- measuring and calculating in longitudinal direction respectively with a series of candidate time cycle
Method is predicted and calculates its error, that minimum time cycle of error is used for as the parameter of laterally longitudinal prediction algorithm pre-
Survey, i.e., according to the different time cycles, laterally longitudinal prediction algorithm is become into multiple fallout predictors, according to last prediction knot
Really, current optimal time cycle corresponding fallout predictor is selected.
Specifically, the candidate time cycle is respectively 6 hours, 12 hours, 1 day and 1 week.
The characteristics of present invention is directed to server performance, is carried out long-term pre- by laterally longitudinal prediction algorithm to behavior pattern
Survey.It is difficult to prediction for time series data and searches out relativity problem, data by is cut into multiple subsequences by the present invention
Data find data dependence.By the correlation of this data, data are entered with horizontal algorithm using longitudinal algorithm respectively
Row prediction, and the result of prediction is adjusted, further improve prediction finally by the periodicity of dynamic adjusting data
Accuracy and reliability.
Disclosed above is only several specific embodiments of the present invention, but the present invention is not limited to this, any this area
Technical staff can think change, should all be within the scope of the present invention.
Claims (5)
1. a kind of DCE performance prediction method based on laterally longitudinal information integration, it is characterised in that including with
Lower step:
Step 1:One long time series data is cut into the Time Sub-series of multiple equal lengths, the length of each Time Sub-series
Spend for T;
Step 2:The period of the day from 11 p.m. to 1 a.m to all generations of step 1 is predicted using Exponential Smoothing curved line arithmetic using longitudinal prediction algorithm
Between sequence be predicted, calculate the relevance between cycle and cycle, obtain longitudinal prediction data and pre- to obtained longitudinal direction
Data are surveyed to be integrated;
Step 3:Indulging that relation pair step 2 between two time points within the time cycle is obtained is passed through using lateral prediction algorithm
It is predicted to prediction data, specifically finds the contact of the data in the cycle, build relational model, to the week all times of searching
The horizontal relationship existed in phase is stored, and then enters line period interior prediction, weighted superposition predicts the outcome, and passes through feedback system
Horizontal relationship is adjusted, is finally predicted the outcome;
Step 4:The horizontal longitudinal prediction algorithm of dynamic integrity
Correspondence different machines different time sections, cycle time relevance is not fixed situation using dynamic laterally longitudinal direction prediction
Algorithm, be specially:
Assuming that prediction length in future is relatively, suffered time cycle influence is the same, in each prediction, by history number
According to last identical prediction length is removed, then call horizontal longitudinal prediction algorithm pre- respectively with a series of candidate time cycle
Survey and calculate its error, be used for predicting using that minimum time cycle of error as the laterally parameter of longitudinal prediction algorithm, i.e.,
According to the different time cycles, laterally longitudinal prediction algorithm is become into multiple fallout predictors, predicted the outcome according to last, to select
Select current optimal time cycle corresponding fallout predictor.
2. the DCE performance prediction method according to claim 1 based on laterally longitudinal information integration, its
It is characterised by, historical data is specially converted into multigroup Time Sub-series by step 1 according to period of time T, wherein:
Historical data is:{x1,x2,x3,...,xn, length is n;
The length of each Time Sub-series is T, is cut into n/T Time Sub-series, i-th of subsequence SiIt is expressed as { xi1,xi2,
xi3,…,xiT, then historical data is cut into<S1,S2,…,Si,…,SL>, wherein L=n/T:
Wherein, x1,x2,x3,...,xnIt is historical data, SiIt is Time Sub-series,
xi1,xi2,xi3,…,xiTThe data of Time Sub-series after cutting.
3. the DCE performance prediction method according to claim 2 based on laterally longitudinal information integration, its
It is characterised by, step 2 comprises the following steps:
Step 21:Long-term forecast, prediction are carried out to the time series of all generations of the first step using Exponential Smoothing curved line arithmetic
Length be T step, predict the outcome as follows:
Wherein,It is longitudinal prediction data;
Step 22:Consolidated forecast data, form prediction data, often by step 21 prediction data point according to the sequence combination of original sampled point
The direction of one fallout predictor prediction be it is longitudinal, data finally splicing be it is horizontal,
For horizontal prediction data, length is T.
4. the DCE performance prediction method according to claim 3 based on laterally longitudinal information integration, its
It is characterised by, step 3 comprises the following steps:
Step 31:Calculate the horizontal relationship between same all midcycle datas
Horizontal relationship refers to the relation between two time points within the time cycle, specific as follows:
Assuming that two data points are x, y, definition y=bx+a is the horizontal association between x, y, wherein, a, b is parameter;
A, b parameter are directly obtained at linear regression, i.e., linear regression method least square method formula is as follows:
Q=(y1-bx1-a)2+(y2-bx2-a)2+…+(yn-bxn-a)2
Minimize Q, first derivation show that design factor formula is as follows:
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>b</mi>
<mo>=</mo>
<mfrac>
<mrow>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mover>
<mi>y</mi>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mo>=</mo>
<mfrac>
<mrow>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mi>n</mi>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mover>
<mi>y</mi>
<mo>&OverBar;</mo>
</mover>
</mrow>
<mrow>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msup>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mn>2</mn>
</msup>
<mo>-</mo>
<mi>n</mi>
<msup>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mo>,</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>a</mi>
<mo>=</mo>
<mover>
<mi>y</mi>
<mo>&OverBar;</mo>
</mover>
<mo>-</mo>
<mi>b</mi>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mo>.</mo>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Wherein, Q is predicted value and the error of calculated value,For sample point x average,For sample point y average;
It whether there is horizontal relationship for two data points, pass through the linear regression parameters come out to cycle training at the same time
Checking, if its error is less than certain threshold values, then it is assumed that there is horizontal relation;Otherwise, think if error exceedes threshold values
Horizontal association is not present before the two time points;
Step 32:The horizontal relationship existed in all time cycles is found, and is stored it;
Step 33:Predict the outcome weighted superposition
After the horizontal relationship of all presence has been searched for by the predicted value of longitudinal algorithm with the predicted value of horizontal algorithm to weight
Mode is superimposed, and obtains prediction weighted results as follows:
Wherein:
It is the prediction weighted results of the i-th row jth row;
w1With w2It is the weight of laterally and longitudinally algorithm respectively;
It is the result of horizontal algorithm prediction;
It is that longitudinal algorithm predicts the outcome;
Step 34:Data are adjusted by horizontal relationship
Horizontal relationship is applied not only to prediction, is also used for adjustment finally to data, it is assumed that two time points of A, B, which exist, laterally closes
System is y=ax+b, wherein, A corresponds to x, and B corresponds to y, and a, b is known parameters, also, obtains A, B by prediction above
Prediction data is xp, yp, then,
Set up cost function:
Q(y-yp)2+(x-xp)2=(ax+b-yp)2+(x-xp)2
It is exactly that two data points meet horizontal relationship that Cost function, which are minimized, carries out first derivation and is set to 0:
2a(aX+b-yp)+2(X-xp)=0
<mrow>
<mi>X</mi>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>x</mi>
<mi>p</mi>
</msub>
<mo>-</mo>
<mi>a</mi>
<mrow>
<mo>(</mo>
<mi>b</mi>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>p</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<mn>1</mn>
</mrow>
</mfrac>
</mrow>
It is X, aX+b respectively to obtain new prediction data point A, B.
5. the DCE performance prediction method according to claim 1 based on laterally longitudinal information integration, its
It is characterised by, the candidate time cycle in step 4 is respectively 6 hours, 12 hours, 1 day and 1 week.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410198278.4A CN104010029B (en) | 2014-05-12 | 2014-05-12 | DCE performance prediction method based on laterally longitudinal information integration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410198278.4A CN104010029B (en) | 2014-05-12 | 2014-05-12 | DCE performance prediction method based on laterally longitudinal information integration |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104010029A CN104010029A (en) | 2014-08-27 |
CN104010029B true CN104010029B (en) | 2017-09-08 |
Family
ID=51370520
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410198278.4A Active CN104010029B (en) | 2014-05-12 | 2014-05-12 | DCE performance prediction method based on laterally longitudinal information integration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104010029B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106201718A (en) * | 2016-07-05 | 2016-12-07 | 北京邮电大学 | A kind of cloud computing resources dynamic retractility method based on load estimation |
CN108241864A (en) * | 2016-12-26 | 2018-07-03 | 摩根士丹利服务集团有限公司 | Server performance Forecasting Methodology based on multivariable grouping |
CN108256893A (en) * | 2016-12-29 | 2018-07-06 | 北京国双科技有限公司 | The analysis method and device of advertisement delivery effect |
CN108200473B (en) * | 2018-02-01 | 2020-12-08 | 深圳创维-Rgb电子有限公司 | CPU power control method, smart television and storage medium |
CN110928634B (en) * | 2018-09-19 | 2023-04-07 | 阿里巴巴集团控股有限公司 | Data processing method, device and equipment |
CN116186017B (en) * | 2023-04-25 | 2023-07-28 | 蓝色火焰科技成都有限公司 | Big data collaborative supervision method and platform |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103544362A (en) * | 2013-11-04 | 2014-01-29 | 国网上海市电力公司 | Harmonic medium and long term prediction method based on two-dimensional curve prediction |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7373287B2 (en) * | 2004-08-09 | 2008-05-13 | Bridgestone Firestone North American Tire, Llc | Finite element analysis tire footprint smoothing algorithm using multiple load cases |
FR2935791B1 (en) * | 2008-09-05 | 2010-09-17 | Thales Sa | SYSTEM AND METHOD FOR FUSION OF METEOROLOGICAL DATA PREDICTED AND MEASURED ON AIRCRAFT |
-
2014
- 2014-05-12 CN CN201410198278.4A patent/CN104010029B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103544362A (en) * | 2013-11-04 | 2014-01-29 | 国网上海市电力公司 | Harmonic medium and long term prediction method based on two-dimensional curve prediction |
Non-Patent Citations (4)
Title |
---|
《A pattern fusion model for multi-step-ahead CPU load prediction》;Dingyu yang 等;《The Journal of systems and Software》;20121227;第1257-1266页 * |
《分布式环境中的性能预测方法》;付继文;《中国优秀硕士学位论文全文数据库信息科技辑》;20130715(第7期);第I137-137页 * |
《时间序列预测中提高预测准确度的几种方法》;戚成功;《预测》;19900430;第9卷(第4期);第47-49页 * |
《电力负荷与电量组合式预测模型及其应用》;赵海青 等;《水电能源科学》;19961231;第14卷(第4期);第275-280页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104010029A (en) | 2014-08-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104010029B (en) | DCE performance prediction method based on laterally longitudinal information integration | |
CN106549772B (en) | Resource prediction method, system and capacity management device | |
CN108846517B (en) | Integration method for predicating quantile probabilistic short-term power load | |
CN103257921B (en) | Improved random forest algorithm based system and method for software fault prediction | |
CN109587713A (en) | A kind of network index prediction technique, device and storage medium based on ARIMA model | |
CN105760213B (en) | The early warning system and method for resources of virtual machine utilization rate under cloud environment | |
CN111198808B (en) | Method and device for predicting performance index, storage medium and electronic equipment | |
CN109272169A (en) | Traffic flow forecasting method, device, computer equipment and storage medium | |
CN106598822B (en) | A kind of abnormal deviation data examination method and device for Capacity Assessment | |
Lujic et al. | Efficient edge storage management based on near real-time forecasts | |
CN103176974A (en) | Method and device used for optimizing access path in data base | |
CN106980874B (en) | A kind of multi-time Scales dimension data fusion method towards distribution big data | |
CN104217091A (en) | Website page view prediction method based on historical tendency weights | |
CN107015900A (en) | A kind of service performance Forecasting Methodology of video website | |
CN104239963A (en) | Method for finding abnormal electric energy meter based on gray GM (1, 1) model | |
CN109492826A (en) | A kind of information system operating status Risk Forecast Method based on machine learning | |
CN104407688A (en) | Virtualized cloud platform energy consumption measurement method and system based on tree regression | |
CN104778185A (en) | Determination method for abnormal SQL (structured query language) statement and server | |
CN106209967A (en) | A kind of video monitoring cloud resource prediction method and system | |
Kantere et al. | Predicting cost amortization for query services | |
US20070233532A1 (en) | Business process analysis apparatus | |
CN109816157A (en) | Project plan optimization method, device, computer equipment and storage medium | |
CN106779147B (en) | Power load prediction method based on self-adaptive hierarchical time sequence clustering | |
CN116244069A (en) | Capacity expansion and contraction method and device, electronic equipment and readable storage medium | |
CN113837383B (en) | Model training method and device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20200116 Address after: Room 1709, Building No. 8, Binjiang West Road, Jiangyin City, Wuxi City, Jiangsu Province Patentee after: Jiangyin Daily Information Technology Co., Ltd. Address before: 200240 Dongchuan Road, Shanghai, No. 800, No. Patentee before: Shanghai Jiaotong University |
|
TR01 | Transfer of patent right |